Self-adaptive bio-inspired optimisation of machine learning pipelines for rejection risk prediction in kidney transplant patients

Kidney transplantation is the treatment of choice for end-stage kidney disease. Kidney allograft rejection is one of the primary mechanisms of graft loss following organ transplantation. The histopathological diagnosis of kidney transplant rejection is complicated by a lack of disease-specific lesions, highlighting the urgent need for more quantitative approaches. Gene expression profiling of graft biopsies can provide evidence of rejection before a clinical phenotype becomes apparent. This thesis aims to improve the prediction of kidney transplant rejection using gene expression microarray data by leveraging bio-inspired optimisation techniques to improve classification performance. We used bio-inspired optimisation algorithms to perform numerical hyperparameter optimisation of classification pipelines trained on biopsy and blood samples of kidney transplant patients. Our results demonstrate that the optimised pipelines improved the predictive capabilities in many cases, particularly when adopting a particle swarm optimisation approach. The genes identified by the optimised pipelines as being important for predictive performance were relevant to the context of transplantation with a diverse immunological scope being independently selected by the different classification pipelines, underscoring the complexity of the immunological response underpinning kidney allograft rejection.

Publisher

University of Galway

Rights

Attribution-NonCommercial-NoDerivatives 4.0 International

cbn

Collections

University of Galway Theses (PhD Theses)